MIT "AI can already replace 11.7% of the U.S. workforce"

Up to $1.2 trillion in wage impact in finance, healthcare, etc.

Simulation jointly conducted by MIT and Oak Ridge National Laboratory

Tennessee, North Carolina, Utah among users of the policy tool

A study released on the 26th (local time) by the Massachusetts Institute of Technology (MIT) analyzed that artificial intelligence (AI) has already reached a level capable of replacing 11.7% of the U.S. workforce. This corresponds to roughly $1.2 trillion in annual wage exposure, centered on the finance, healthcare, and professional services sectors.

The study was conducted based on the labor simulation tool "Iceberg Index," jointly developed by MIT and Oak Ridge National Laboratory (ORNL). The index models how AI affects occupations and regions across 151 million workers in the United States.

Co-principal investigator and ORNL director Prasanna Balaprakash said, "We are building a digital twin of the U.S. labor market," adding, "We can precisely identify which tasks AI can currently perform." A digital twin refers to a virtual model that mirrors the real structure, elements, and data of reality.

The Iceberg Index derives the skills and occupations AI can immediately perform based on △923 occupations △32,000 skills △3,000 counties (U.S. local government unit) information. The researchers noted that the visible layoffs and occupational transitions in technical, computer, and IT fields account for only 2.2% (about $211 billion) of total wage exposure.

Beneath the "surface," however, there exists about $1.2 trillion worth of potentially replacable tasks centered on routine work such as human resources, logistics, finance, and administrative office tasks.

MIT put forward the concept of "exposure hotspots" as a core idea in this analysis. This means that the likelihood of AI replacement and automation is concentrated in certain regions or occupations, creating points where shocks are focused rather than an impact that spreads evenly. MIT explained that the Iceberg Index can identify risk distributions down to ZIP code and census block levels.

The researchers clarified that the Iceberg Index is not designed to predict exactly when or where specific jobs will disappear. Instead, they emphasized that it is a tool to accurately assess current exposure at precise skill and occupational levels based on current AI capabilities and to pre-test policy scenarios.

Three states — Tennessee, North Carolina, and Utah — used their own labor data to validate the model and have begun using the tool for actual policy design. Tennessee officially cited the Iceberg Index in its "AI Workforce Action Plan" released this month.

North Carolina state senator Diandria Salvador said, "Using this tool, you can drill down to census block levels in a specific county to analyze which skills are being performed and the degree of automation potential," adding, "You can also assess impacts on regional GDP and employment."

The Iceberg Index also overturns the assumption that AI adoption risk would be concentrated only in tech jobs on the West Coast, such as Silicon Valley, and in East Coast cities like New York and Boston. The study found exposure across all 50 states, especially in inland and rural areas that have been sidelined in AI discussions.

To address this, the research team built an interactive policy experimentation environment allowing each state government to simulate allocations of workforce training funds, adjustments to retraining programs, changes in the speed of technology adoption, and more.

The report evaluated, "Project Iceberg enables policymakers and business leaders to identify exposure hotspots, set priorities for workforce retraining and infrastructure investment, and test intervention strategies before investing actual billions of dollars."

Director Balaprakash, discussing talks with the Tennessee state government, said, "Tennessee's core industries — healthcare, nuclear, manufacturing, and transportation — still have high shares of physical labor, making full replacement through digital automation alone difficult," and added, "We need to explore ways to strengthen the industrial base using robotics and AI."

The research team defined this index not as a finished product but as a "policy-experiment sandbox" to prepare for AI shocks. Senator Salvador said, "This tool is optimized for running various scenarios directly and formulating preparedness strategies."

New York = Shin-young Park, correspondent nyusos@hankyung.com